BACKGROUNDThe objective of the current study was to evaluate the reproducibility of the Fuhrman nuclear grading system as well as its independent predictive value in a series of patients with conventional renal cell carcinoma (RCC).METHODSThe authors selected 388 patients who had undergone surgical treatment for conventional RCC between 1986 and 2000. Pathology slides from the selected patients were reviewed by a single pathologist, who reassigned a Fuhrman nuclear grade and assessed the presence of tumor necrosis. The pathologist was blinded to both the original pathologic diagnosis and follow‐up data. The κ statistic was used to evaluate concordance between original and reviewed nuclear grades. The log‐rank test was used for univariate analyses, and a Cox proportional hazards model was used for multivariate analyses.RESULTSThe original Fuhrman nuclear grade was Grade 1 (G1) in 111 patients (28.6%), G2 in 141 patients (36.3%), G3 in 108 patients (27.8%), and G4 in 28 patients (7.3%). After pathology slide review, nuclear grades were reassigned as follows: G1 in 49 patients (12.6%), G2 in 138 patients (35.6%), G3 in 150 patients (38.7%), and G4 in 51 patients (13.1%). The grade of concordance was moderate (κ = 0.44; P < 0.001). Univariate analyses identified three separate prognostic categories defined by nuclear grade (G1 and G2 vs. G3 vs. G4). Both the original and the reviewed Fuhrman nuclear grading systems were capable of independently predicting disease‐specific survival in patients with conventional RCC.CONCLUSIONSThe interobserver reproducibility of Fuhrman nuclear grading was moderate. The substantial overlap in survival curves for G1 and G2 tumors provided an opportunity to cluster those categories, and the resulting three‐tiered nuclear grading system was an independent predictor of cause‐specific survival in patients with conventional RCC. Other independent predictors of survival included pathologic stage and tumor necrosis status. Cancer 2005. © 2004 American Cancer Society.